Out of curiosity, let’s see many points Scott Alexander would get for the last three months of 2025. Skipping the open threads and links and guest posts and highlights from comments and meetup announcements and contests and grant results...
Well, yes. The question is, how much exactly. I mean, what are the points even supposed to reflect?
The thing I am trying to capture is “value for the reader”. As a reader of ACX, I prefer longer articles to shorter articles, assuming constant frequency. But I prefer two articles a week to one twice-as-long article.
Mathematically speaking, this means that the function “how many words for N points” should be growing, faster than linearly. But that still leaves many options. Intuitively, I chose 250n²+250n, as a quadratic (faster than linear) expression with backwards compatibility (results in 500 for n=1). If we extrapolate that further, it would be 5 points for 7.5k words, 6 points for 10.5k words, 7 points for 14k words, 8 points for 18k words, 9 points for 22.5k words, and 10 points for 27.5k words.
According to this extended scale, Scott would get 6+4+1+3+10+2+2+3+1+2+1+2+3+3+2+3+1+2= 51 points.
But that assumes that the particular quadratic function is the right one. If we chose an exponential function instead: 1 point for 500 words, 2 points for 1k words, 3 points for 2k words, 4 points for 4k words, 5 points for 8k words, and 6 points for 16k words, that would give slightly more points to articles below 4k, but fewer points above 10k, together 5+4+2+3+6+2+3+3+2+3+1+3+3+4+2+4+2+3= 55 points.
Both functions seem to pass the smell test—Scott should get more than 30 points, but not like an order of magnitude more (especially not after I have filtered out the links and contests and highlights). Getting about twice as much sounds about right. Still doesn’t answer which function to choose; both seem okay precisely because they give similar results.
Another question is, how often will Halfhaven participants produce articles longer than 5k words. I know I probably won’t, which makes the scoring of articles over 5k words seem irrelevant. I am open to changing my mind if someone actually starts writing the articles of such length, but at this moment it feels like if someone can write that kind of article repeatedly, they no longer need this project.
tl;dr—the choice is arbitrary, and probably irrelevant for Halfhaven participants, so although I kinda agree with you, I am not going to change it now (but might change it in April, dunno)
Out of curiosity, let’s see many points Scott Alexander would get for the last three months of 2025. Skipping the open threads and links and guest posts and highlights from comments and meetup announcements and contests and grant results...
What Is Man, That Thou Art Mindful Of Him? (10k) 🪙🪙🪙🪙
Book Review: If Anyone Builds It, Everyone Dies (6k) 🪙🪙🪙🪙
Defining Defending Democracy: Contra The Election Winner Argument (1.3k) 🪙
Sources Say Bay Area House Party (3.3k) 🪙🪙🪙
The Fatima Sun Miracle: Much More Than You Wanted To Know (28k) 🪙🪙🪙🪙
Fascism Can’t Mean Both A Specific Ideology And A Legitimate Target (1.7k) 🪙🪙
Tech PACs Are Closing In On The Almonds (2.5k) 🪙🪙
Model City Monday 10/27/25 (3.1k) 🪙🪙🪙
Writing For The AIs (1.2k) 🪙
The Bloomer’s Paradox (2.7k) 🪙🪙
In What Sense Is Life Suffering? (800) 🪙
What Happened To SF Homelessness? (2.3k) 🪙🪙
The New AI Consciousness Paper (3.8k) 🪙🪙🪙
Why AI Safety Won’t Make America Lose The Race With China (4.4k) 🪙🪙🪙
The Good News Is That One Side Has Definitively Won The Missing Heritability Debate (1.7k) 🪙🪙
Vibecession: Much More Than You Wanted To Know (4.2k) 🪙🪙🪙
The Pledge (1.1k) 🪙
Against Against Boomers (2.1k) 🪙🪙
Total: 43 points.
Surely 28K should be at least 5 points!
Well, yes. The question is, how much exactly. I mean, what are the points even supposed to reflect?
The thing I am trying to capture is “value for the reader”. As a reader of ACX, I prefer longer articles to shorter articles, assuming constant frequency. But I prefer two articles a week to one twice-as-long article.
Mathematically speaking, this means that the function “how many words for N points” should be growing, faster than linearly. But that still leaves many options. Intuitively, I chose 250n²+250n, as a quadratic (faster than linear) expression with backwards compatibility (results in 500 for n=1). If we extrapolate that further, it would be 5 points for 7.5k words, 6 points for 10.5k words, 7 points for 14k words, 8 points for 18k words, 9 points for 22.5k words, and 10 points for 27.5k words.
According to this extended scale, Scott would get 6+4+1+3+10+2+2+3+1+2+1+2+3+3+2+3+1+2= 51 points.
But that assumes that the particular quadratic function is the right one. If we chose an exponential function instead: 1 point for 500 words, 2 points for 1k words, 3 points for 2k words, 4 points for 4k words, 5 points for 8k words, and 6 points for 16k words, that would give slightly more points to articles below 4k, but fewer points above 10k, together 5+4+2+3+6+2+3+3+2+3+1+3+3+4+2+4+2+3= 55 points.
Both functions seem to pass the smell test—Scott should get more than 30 points, but not like an order of magnitude more (especially not after I have filtered out the links and contests and highlights). Getting about twice as much sounds about right. Still doesn’t answer which function to choose; both seem okay precisely because they give similar results.
Another question is, how often will Halfhaven participants produce articles longer than 5k words. I know I probably won’t, which makes the scoring of articles over 5k words seem irrelevant. I am open to changing my mind if someone actually starts writing the articles of such length, but at this moment it feels like if someone can write that kind of article repeatedly, they no longer need this project.
tl;dr—the choice is arbitrary, and probably irrelevant for Halfhaven participants, so although I kinda agree with you, I am not going to change it now (but might change it in April, dunno)